Describe the bug
My ChatGPT Workspace Agent (Business plan, research preview) intermittently gets stuck in a loop and starts returning the exact same response over and over. Once it enters this state, the output is identical regardless of what I send next — the agent keeps repeating its previous message and does not produce a new answer until I reset the thread.
This happens occasionally during normal use. It is not tied to any specific prompt or workflow, and I have not found a reliable way to reproduce it on demand.
Environment
- Plan: ChatGPT Business (Workspace Agents, research preview)
- Surface: ChatGPT (not Slack)
- Model: GPT-5.5
- Reasoning effort: medium
Behavior
- Expected: the agent produces a fresh, relevant response per turn and ends the turn normally.
- Actual: at random points the agent emits an identical response repeatedly and cannot break out of the loop on its own. Whatever I type next, it returns the same text. Starting a new thread clears the state.
Frequency / reproducibility Intermittent and unpredictable — it surfaces sporadically during regular work. I can’t give a reliable repro or a precise frequency, and I don’t have a saved session or screenshots from the occurrences so far.
Questions
- Is this a known issue with Workspace Agents (or the underlying model not emitting a termination signal at the end of a turn)?
- Is there a recommended mitigation while the feature is in research preview?
- If it would help the team, what’s the best way to capture diagnostics the next time it happens — should I grab the session ID / use
/feedbackfrom within that thread?